E-Ink News Daily

← Back to list

Underdog shows off a 32MB speech model in new demo

Underdog AI has released Woof, a 32MB open-source text-to-speech model with 31 million parameters that can run entirely offline on local devices. The company showcased it through an interactive browser-based game where users type text or adjust a tuner to generate speech and music in real time. While too small to replace large-scale cloud models, it targets local assistant experiences and personalized content generation.

Background

There is a growing trend toward smaller, more efficient AI models that can run locally on consumer devices, reducing reliance on cloud infrastructure. Companies like TinyCat, Mobvoi, and others have been pursuing similar on-device speech and language model approaches.

Source
TechCrunch
Published
Nov 6, 2026 at 09:02 PM
Score
5.0 / 10